Good models don't save bad architecture, and most teams learn that the hard way. The post Most AI Agents Fail in Production Because They’re Built Backwards appeared first on Towards Data Science.

Discussion | Link

Google is folding Display Ads into its AI-powered Demand Gen platform, marking the end of a long-standing digital advertising model. The Google Display Network (GDN) has been a staple of the open internet for almost twenty years. Marketers previously relied on its predictable framework to target placements, bid on audiences, and A/B test static creative […] The post Google folds Display Ads into AI-first Demand Gen platform appeared first on AI News.

Moving data from your operational database has traditionally meant setting up and...

🧠 A new tool grounded in over 25 frameworks like the Kano Model, Jobs to be done, Pre-mortems, Porter’s Five Forces and more.

How deepfakes rocked a high school; BusPatrol giving AI camera data to cops; and a big time crash out.

Generative artificial intelligence (Gen AI) has been the most disruptive force to hit education in a long time, and educators are still absorbing the impact. My colleagues and I have had to reprogram our approaches and classroom assessments as Gen AI infiltrates every aspect of the academic setting. I was not ready to have Gen AI affect my work with Holocaust education. I’ve come to believe that the risk of harmful outcomes using Gen AI to recreate images or scenes from the Holocaust outweigh the potential educational benefits. And I’m not alone. In a 2024 report titled “AI and the Holocaust: rewriting history?”, UNESCO warned that AI could be used to fabricate realistic images or videos casting doubt on whether the Holocaust occurred — a phenomenon that could fuel greater antisemitism. And in 2025, the Auschwitz-Birkenau Memorial Museum argued that using AI “to generate fictional images of Auschwitz victims […] is not a tribute. It is a profound act of disrespect to the memory of those who suffered and were murdered” and that “it undermines the integrity of historical truth.” The ethics of duplication The first time I encountered Gen AI in Holocaust education was in Joe Berlinger’s 2024 six-episode documentary, Hitler and the Nazis: Evil on Trial, where original news broadcast recordings of American journalist William L. Shirer were used to generate an approximation of his voice to read out the diaries he wrote when reporting from Nazi Germany. Gen AI can be beneficial in certain circumstances. For example, the use of Shirer’s voice can help to better engage learners. With education and increased awareness as the main motives for employing Gen AI, the Jewish Children’s Museum in Brooklyn has used short videos to present Holocaust history to elementary school students in an age-appropriate manner. Algorithms and AI researchers Atay Kozlovski and Mykola Makhortykh assert in “Digital Dybbuks and Virtual Golems: The Ethics of Digital Duplicates in Holocaust Testimony” that there is a human responsibility to interact with AI ethically. They state: “This concern closely relates to the harm-risk mitigation condition, which demands identifying and evaluating potential risks and requires proactive steps to reduce them, whether through design or regulation.” Local Journalism Initiative reporter Maggie Macintosh of the Prince Albert Herald reported in 2025 that “the Canadian Museum for Human Rights is using artificial intelligence to introduce visitors to lifelike avatars of aging survivors of the Holocaust who can hold real-time conversations with them.” This would seem to suggest that Canada’s approach to contemporary Holocaust education embraces Gen AI. However, citing online content manipulated by Gen AI to depict Hitler spouting antisemitic tirades, but in English instead of German, has been described by many scholars as a “worrying trend.” Clickbait, denial and the feed The most common place to encounter Holocaust-related Gen AI is on social media. Although the average consumer is aware they must be critical of the images and videos they consume online, even the most savvy scrollers can still absorb misinformation. When people come across material related to the Holocaust on social media, sometimes it can be immediately obvious that it’s AI-generated. Some material might be expressly identified, but ambiguity could lead people to question the authenticity of not just the representation, but history itself. Mindblown AI, a digital content creation platform that focuses on AI-generated historical storytelling, produced a short clip first posted to Facebook in 2021 that illustrates the short life of Anne Frank. The sequence depicts an AI approximation of Frank at various stages of her childhood leading up to her murder. It’s also set to ABBA’s catchy song “Lay All Your Love on Me.” Mindblown AI published a similar video montage in 2024 about Irma Grese, a convicted war criminal and guard at concentration camps Ravensbrück, Auschwitz-Birkenau and Bergen-Belsen. Female Nazi concentration camp guards: the true horror lies in their similarities to ourselves Another, more alarming AI-generated video by History Fact Untold, a YouTube channel that publishes short-form historical trivia, details Grese’s violence and depravity. It shows a sexualized Grese walking through huddled masses of emaciated prisoners. The content is not intentionally misleading; in fact the publisher is transparent about the use of Gen AI. But if a viewer fails to situate these videos within the realm of recreated historical representation rather than authentic documentation, there is room to interpret the Holocaust as less destructive than it was. Gen AI photos depicting the Holocaust are arguably more problematic for audiences because high-quality renderings may appear quite convincing. DW News, a German news and current affairs channel, highlights one such image of an emaciated man playing the violin passionately while fellow prisoners are marched to his rhythm along their path to execution. DW News analysis cites several visual indicators that reveal the image is fake: three non-linear strings on the violin, inaccuracies in attire including prisoners wearing belts and the number of fingers on background figures. A teaching opportunity? A student recently asked me whether using Gen AI to create images to accompany documented testimonies might be an effective way of engaging this generation of learners. That’s an interesting point. We were, after all, discussing graphic literature and Holocaust memory narration. But at the end of the day, it’s very much a slippery slope. In academia, we are trained to be vigilant. Instilling this same vigilance in future generations is becoming more critical than ever. We can accept this as an opportunity to actively engage AI-savvy young people in combating revisionism, distortion and denial. Regan Lipes does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Discussion | Link

The foreign exchange market is really moving away from pure intuition and toward a space shaped by speed, data and precision. By using automated systems in your routine, you can approach volatility with a level of discipline that manual trading often struggles to maintain. Every entry and exit can be based on clear rules, not […] The post Exploring the Benefits of AI Bots for Forex Trading in Forex Markets appeared first on AI News.

Police clear the campus of Brooklyn College on May 8, 2025, after students established an encampment to protest the Gaza war. Michael Nigro/Pacific Press/LightRocket via Getty Images Younger Americans have soured on the second Donald Trump presidency, but they are not protesting it. Despite an unpopular Iran war and an even more unpopular Trump administration, college campus protests nationwide have gone silent. And at many schools, student activism is virtually nonexistent. This silence comes in the wake of a relentless Trump administration war on campus speech that has involved lawsuits, arrests, deportations and expulsions. Reports cite a range of complicated factors for the restraint, from apathy to technology-induced incapacity. But as public policy and law and social science experts, we believe students aren’t protesting for a very simple reason: They are afraid. They are self-censoring and disengaging from campaign activism to avoid punitive measures. In law and social science, we call this impact a chilling effect – the behavioral tendency for people in face of a threat to self-censor and restrain their activities for self-protection. It’s increasingly clear to us that these impacts are not incidental or ancillary to Trump administration policy. Rather, the chilling effects are the point. This is the closest thing to a consistent governing strategy in Trump’s second term. The broader chill of Trump threats Chilling effects can be subtle, but today they are everywhere. And it’s not just students who are chilled by Trump administration threats. Professors are censoring themselves in lectures and rewriting syllabuses. Researchers are stripping grant applications of words that might attract federal scrutiny, or abandoning the topics entirely. Media outlets are modifying their news coverage to avoid Trump lawsuits or sanctions. Law enforcement and regulatory agencies are refusing to investigate Trump-aligned actors inside or outside government, and major national law firms are declining cases challenging Trump administration policies. Publishers are “stepping back” from LGBTQ+ books and other progressive subjects. Many in targeted immigrant communities are afraid to leave home to go to work or school. In most cases, these people and institutions are not being specifically targeted or threatened by Trump. But they are afraid, and their fear is doing the administration’s work for it. They stay silent, avoid attention and confrontation, and look the other way. In other cases, they change their speech and behavior to accommodate or conform to the administration’s worldview. Of course, there are counterexamples, such as the winter protests in Minneapolis in response to brutality by agents with U.S. Immigration and Customs Enforcement, and the recent “No Kings” rallies. But even here, the broader but less visible trend – chilling effects – is evident. Minneapolis police officers arrest and scatter protesters on the campus of the University of Minnesota in Minneapolis on Feb. 5, 2026. Richard Tsong-Taatarii/The Minnesota Star Tribune via Getty Images For instance, in recent reporting on the latest No Kings rallies, many media outlets observed that students were noticeably missing, despite the Trump administration’s unpopularity among younger Americans. A persistent strategy We believe none of this is by accident. In a new book, “Chilling Effects: Repression, Conformity, and Power in the Digital Age,” one of us – Jon Penney – explains how law, technology, and state and corporate power are weaponized to chill and repress, and the dangers this poses for the United States and other democratic societies. The other – Bruce Schneier – has extensively studied the security infrastructure enabling this. What we see isn’t gratuitous government cruelty, chaos or vengeance. Instead, we see a persistent strategy to maximize fear and chilling effects in ways that are corrosive to freedom and democracy. Research suggests that surveillance, personal threats, uncertainty and abuse of power are key factors in doing so. The federal government has a clear and systematic pattern of employing these very mechanisms across a number of domains far beyond campuses. They are evident in militarized raids by Immigration and Customs Enforcement and in journalists being arrested and indicted for reporting on protests. They are made clear in the long list of political enemies the Trump administration has investigated or threatened, including the Federal Reserve chairman. And they can also be seen in the weaponization of technology, including ramping up surveillance to target critics and protestors. Corrosive to freedom and democracy History offers some guidance on impacts. During the McCarthy era, overreaching laws, surveillance, and public and private sector reprisals ostensibly targeted alleged communists. But the real aim was often to suppress progressive journalists, trade unions and political opposition. In the 1960s, these same tactics were reused by Southern states to chill the Civil Rights Movement. Historians have written about how the widespread fear and conformity of these periods reshaped American society in enduring ways, including the destruction of progressive political movements and both delaying and muting the Civil Rights Movement itself. When such state threats are systematized, they can foment a broader climate of fear, self-censorship and conformity. In that climate, dissenting speech, political opposition, democratic mobilization and other checks on power become increasingly difficult, even dangerous. It is no surprise, for instance, that Trump critics regularly admit to self-censorship, fearing for their safety. Chilling effects are thus not only repressive – causing self-censorship – but productive. They produce conforming and compliant speech and behavior, which can have longer-term social impacts. They not only undermine protected rights and suppress accountability but can promote social change – even without a popular mandate to do so. University of Chicago police patrol the campus after dismantling a pro-Palestine encampment on May 7, 2024. Alex Wroblewski/AFP via Getty Images This latter point is often missed. It explains Trump’s assaults on universities and cultural institutions such as the Kennedy Center for the Arts and the Smithsonian. Often dismissed as peculiar Trump obsessions, they are fully consistent with Project 2025 – the sweeping policy blueprint for Trump’s second term authored by a coalition of conservative groups and its call to target the “institutions of American civil society” and “wield federal power” to “reverse” decades of progressive cultural advancements. In the near term, this means an increasingly weakened democratic society, with the government and its patrons enjoying freedom to pursue their objectives. Over the long term, this can mean a changed society as more conformist and compliant speech and culture become more widely accepted and entrenched. Not inevitable In our view, this future is not inevitable, just as the McCarthy era “Red Scare” and violent civil rights era repression were not. In both cases, fear and chilling effects were resisted in law and civil society, as they can be today. But the central mechanisms – surveillance, uncertainty, personal threats and abuse of power – would need to be addressed. For instance, new legislation could ensure justice for lawless government actors and constrain surveillance. Courts can block abuses of federal power, including illegal arrests, detentions and mass citizen databases. The media, lawyers and civil society can hold the government accountable. And students, teachers, universities and cultural institutions can resist the tendency to self-censor and conform. The citizen mobilization in Minnesota and the No Kings rallies are examples of that. But to resist chilling effects and their dangers over the long term, this would have to be the norm, not the exception. Jon Penney has acted as an expert witness, on a pro bono basis, for the American Civil Liberties Association in litigation challenging the legality of government surveillance. Bruce Schneier does not work for, consult, own shares in or receive funding from any company or organization that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Surveys and polls help societies understand what people think about issues in politics, health, education and much more. But fewer people these days tend to respond, so pollsters have to reach out more widely, which raises cost considerably. One survey provider prices a 10 minute survey of 1,000 people in the tens of thousands of dollars. Could AI models stand in for hundreds or thousands of people, emulating the range of answers humans would provide? This practice, known as synthetic surveys or silicon sampling, is already happening, and it’s far less expensive. But are the results trustworthy? I am a machine learning researcher. I study large language models and their uses in medicine and science. These systems change constantly as companies update them. Different prompts, settings and model versions can produce very different answers to questions. That trait can make models difficult to use reliably in social science research, but it can help simulate replies of many humans, what researchers call “synthetic respondents.” To create 10,000 answers from ChatGPT, for example, a pollster would prompt the model with some basic respondent demographics and context, such as “You are a young college-going urban voter with conservative political views. Respond to the following questions.” Researchers can change the demographic settings to elicit many different responses from ChatGPT for the same query. The model also has its own internal randomness, so it naturally generates different replies to the same question asked repeatedly. In this way, researchers can combine prompting and randomness to create 10,000 different synthetic responses. Simulations are not opinions Pollsters have long used statistical models to generalize results from a finite number of replies. And analysts can reach different conclusions from the same survey data. Studies of synthetic respondents suggest they may be even more sensitive than people to small changes in prompts or settings, producing sharply different results. But the use of synthetic respondents raises a deeper issue. Surveys are not just prediction tools. They are measurement tools meant to capture what people actually think. A thermometer measures your temperature directly. You would not trust one that estimated your temperature by consulting an AI model instead. Researchers who poll AI systems instead of people are not measuring public opinion, they are only simulating it. Jose Carlos Cerdeno Martinez via Getty Images Large language models and other AI tools inherit biases and blind spots from the data they train on. For example, AI can oversimplify or distort opinions from groups of people who are underrepresented online. Traditional polling also has biases, but many biases in modern AI systems are hidden from public view inside closed proprietary models. To make matters worse, pollsters may present results from synthetic respondents to the public as if they came from surveys of people. These shortcomings can erode trust in polls and survey research. They also raise an interesting paradox. Synthetic data, created by computers or simulations, is widely used in modern AI. It helps train AI systems for medicine, finance, robotics, self-driving cars and other disciplines. So why do synthetic survey responses seem more problematic? The key difference is that synthetic data is checked against reality. A self-driving car may train on synthetic images and videos of different road conditions, but an automaker would never deploy the car on public roads without extensive real-world testing. If synthetic data hurts performance, engineers can correct, retrain or replace the system. Researchers may treat synthetic survey responses as public opinion itself, but the system is not measuring public opinion. It is running a simulation of public opinion based on data it was trained on. If the simulated opinions distort reality, researchers may not realize it until flawed conclusions have already shaped public policy, business decisions or scientific research. More efficient design and analysis Nevertheless, there are ways AI can help survey research without weakening the measurement of public opinion. AI tools can help survey researchers write clearer questions by simplifying wording, reducing ambiguity and eliminating repetition. They can help avoid unnecessary questions, making it easier for people to respond. These tools can also adapt surveys across languages. Once a survey is done, AI can help researchers organize large volumes of open-ended responses, summarize recurring themes and handle incomplete surveys more efficiently than human analysts. Some researchers are exploring hybrid approaches that combine smaller human surveys with AI-assisted analysis. Decision makers use surveys and polls to listen to and understand the voices of people affected by their decisions. Replacing human respondents with synthetic respondents risks weakening that connection. At the same time, falling response rates and rising costs are real survey challenges. I’m confident that further research can find ways to use AI transparently and effectively, in a scientifically defensible way, without replacing people. Ambuj Tewari receives funding from NSF and NIH.

The cybersecurity industry of 2006 barely resembled today's billion-dollar behemoth. As part of Dark Reading's 20th anniversary celebration, we trace the industry's evolution through a technology lens.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Stay on top of what’s going on in AI this summer Here at MIT Technology Review, we understand exactly how relentless the pace of news from the world of artificial intelligence…

When large language models, or LLMs for short, produce outputs, several criteria are at stake, including not only overall response relevance but also coherence and creativity.

Why good data work gets ignored after delivery. The post They Requested It. I Built It. Nobody Ever Used It. appeared first on Towards Data Science.

Discussion | Link

Discussion | Link

Historically “uninvestable” sectors are getting more opportunities to bring in funding, creating a rare opportunity for specialized founders and investors to disrupt trillion-dollar industrial markets with vertically integrated, AI-driven software. In this guest commentary, Thomas Cuvelier of RTP Global explains how VCs are using AI to overcome legacy barriers.

Pretty little phones with pretty big price tags.

Cisco and OpenAI are redefining enterprise engineering with Codex, helping Cisco scale AI-native development, accelerate AI Defense work, and automate defect remediation.